save_data: /mnt/netapp1/Proxecto_NOS/mt/corpus/iacobus/en-gl/aut/models
## Where the vocab(s) will be written
src_vocab: /mnt/netapp1/Proxecto_NOS/mt/corpus/iacobus/en-gl/aut/models/run/vocab/bpe.vocab.src
tgt_vocab: /mnt/netapp1/Proxecto_NOS/mt/corpus/iacobus/en-gl/aut/models/run/vocab/bpe.vocab.tgt
overwrite: True

# Corpus opts:

data:
    en-gl:
        path_src: /mnt/netapp1/Proxecto_NOS/mt/corpus/iacobus/en-gl/aut/train.en35k.txt
        path_tgt: /mnt/netapp1/Proxecto_NOS/mt/corpus/iacobus/en-gl/aut/train.gl35k.txt
        transforms: [bpe, filtertoolong]

    valid:
        path_src: /mnt/netapp1/Proxecto_NOS/mt/corpus/iacobus/en-gl/aut/valid.en35k.txt
        path_tgt: /mnt/netapp1/Proxecto_NOS/mt/corpus/iacobus/en-gl/aut/valid.gl35k.txt
        transforms: [bpe, filtertoolong]

### Transform related opts:

#### Subword
src_subword_model: /mnt/netapp1/Proxecto_NOS/mt/corpus/iacobus/en-gl/aut/en_35k.code
tgt_subword_model: /mnt/netapp1/Proxecto_NOS/mt/corpus/iacobus/en-gl/aut/gl_35k.code
src_subword_vocab: /mnt/netapp1/Proxecto_NOS/mt/corpus/iacobus/en-gl/aut/models/run/vocab/bpe.vocab.src
tgt_subword_vocab: /mnt/netapp1/Proxecto_NOS/mt/corpus/iacobus/en-gl/aut/models/run/vocab/bpe.vocab.tgt
#src_subword_model: ../sentencepiece/en-gl/en.sp.model
#tgt_subword_model: ../sentencepiece/en-gl/gl.sp.model
src_subword_type: bpe
tgt_subword_type: bpe

src_subword_nbest: 1
src_subword_alpha: 0.0
tgt_subword_nbest: 1
tgt_subword_alpha: 0.0

##embeddings
src_embeddings: /mnt/netapp1/Proxecto_NOS/mt/treino_data/embeddings/en.emb.txt
tgt_embeddings: /mnt/netapp1/Proxecto_NOS/mt/treino_data/embeddings/gl.emb.txt

## supported types: GloVe, word2vec
embeddings_type: "word2vec"

# word_vec_size need to match with the pretrained embeddings dimensions
word_vec_size: 300

#### Filter
src_seq_length: 150
tgt_seq_length: 150

# silently ignore empty lines in the data
skip_empty_level: silent

# General opts
# Update the path to the most recent checkpoint available
#opt.train_from: /home/compartido/ramom/es-gl/models/_step_260000.pt
#train_from: /home/compartido/ramom/es-gl/models/_step_250000.pt
save_model: /mnt/netapp1/Proxecto_NOS/mt/corpus/iacobus/en-gl/aut/models/
keep_checkpoint: 50
save_checkpoint_steps: 10000
average_decay: 0.0005
seed: 1234
report_every: 1000
train_steps: 400000
valid_steps: 10000

# Batching
queue_size: 10000
bucket_size: 32768
world_size: 1
gpu_ranks: [0]
batch_type: "tokens"
#batch_size: 1500
batch_size: 4096
valid_batch_size: 64
batch_size_multiple: 1
max_generator_batches: 2
accum_count: [4]
accum_steps: [0]

# Optimization
model_dtype: "fp32"
optim: "adam"
learning_rate: 2
#learning_rate: 0.00005
warmup_steps: 8000
decay_method: "noam"
adam_beta2: 0.998
max_grad_norm: 0
label_smoothing: 0.1
param_init: 0
param_init_glorot: true
normalization: "tokens"

# Model
encoder_type: transformer
decoder_type: transformer
position_encoding: true
enc_layers: 6
dec_layers: 6
heads: 8
#rnn_size: 512
hidden_size: 512
word_vec_size: 512
transformer_ff: 2048
dropout_steps: [0]
dropout: [0.1]
attention_dropout: [0.1]
share_decoder_embeddings: true
share_embeddings: false